Open nick-youngblut opened 2 months ago
Check out the discussion at Bioconductor/basilisk.utils#11; from what I understand, the miniforge
setup uses mamba
under the hood. The idea is to have this become the default in the next release, assuming that the inevitable struggles on Windows can be resolved.
The idea is to have this become the default in the next release
That's great! Thanks for the info.
inevitable struggles on Windows can be resolved
Does anyone actually do bioinformatics on Windows machines? ...especially when running Linux VMs on a cloud service is so easy and cheap? There's also WSL, so why use Windows?
Does anyone actually do bioinformatics on Windows machines? ...especially when running Linux VMs on a cloud service is so easy and cheap? There's also WSL, so why use Windows?
Apparently 50% of BioC package downloads are for the Windows binaries, as of a few years ago - @vjcitn could give more recent statistics. So, not a negligible proportion of users. I'd guess they're probably bench scientists just trying to get something done with whatever machine they were given; it would be a pretty high barrier to entering bioinformatics if they needed a cloud account (I don't even have one) or set up WSL (I haven't) to get started.
@nick-youngblut Note that you can get faster installation on non-windows machines now if you set the environment variable
BASILISK_MINICONDA_VERSION=py311_23.11.0-2
see section 3.2 of the basilisk vignette. That version of miniconda actually uses mamba resolution IIUC.
To take advantage of this you might have to remove content in the cache at basilisk.utils::getExternalDir()
Thanks @vjcitn for the advice!
To take advantage of this you might have to remove content in the cache at basilisk.utils::getExternalDir()
Yeah, both setting BASILISK_MINICONDA_VERSION
and checking basilisk.utils::getExternalDir()
is definitely a bit of a burden for users.
mamba is substantially faster than conda, and moreover, a micromamba install is smaller and faster to install versus miniconda.
As an example, I'm using the FLAMES bioconductor package, which is using basilisk to create a conda env:
The creation of the conda env and install of the python packages took >15 minutes; whereas with mamba, the install would take ~1-2 minutes.